MemWeaver

What this means for you: - Functionally coherent: The skill does what it says — it reads your agent memory files, runs local Python scripts to collect them, then uses the Agent/LLM to derive patterns and ask you questions. The included scripts do not phone home or require keys. - Privacy risk (important): The scripts are local, but the analysis step requires an LLM. If your agent uses a cloud-hosted model (OpenAI, Anthropic, etc.), your full memory content will likely be sent to that provider. If you want the analysis to remain private, run the skill with a locally hosted model or reduce the data sent (use --days small, remove very sensitive notes before running). - Practical precautions: - Inspect the two Python scripts (collect_memory.py, save_profile.py) yourself — they are small, use only stdlib, and only read/write workspace files. - Run collect_memory.py with a small --days first to see what will be included and review the JSON output before handing it to a model. - Confirm where your Agent's model runs. If using a hosted LLM, do not feed highly sensitive memory to it. - Keep backups or run in an isolated/test workspace if you want to trial the skill. - If you need stronger guarantees, request the author clarify model-hosting assumptions or modify the workflow to limit content before analysis. - No red flags for hidden network endpoints, extra credentials, or unusual install behavior were found. If you do not want your memory analyzed by an external hosted model, do not invoke the skill until you confirm a local-model setup.